app.py

import gradio as gr
import whisper
import os
import tempfile
import numpy as np

# Load Whisper model (this happens once at startup)
print("Loading Whisper model...")
model = whisper.load_model("base")  # Options: tiny, base, small, medium, large
print("Model loaded successfully!")

def transcribe_audio(audio):
    """
    Transcribe audio file using Whisper
    
    Args:
        audio: Can be:
            - File path (string)
            - Tuple of (sample_rate, audio_data) from microphone
            - None
    
    Returns:
        Transcription text and detected language
    """
    if audio is None:
        return "No audio provided", "N/A"
    
    try:
        # Handle different audio input types
        if isinstance(audio, tuple):
            # Microphone input: (sample_rate, audio_data)
            sr, audio_data = audio
            
            # Save to temporary file
            with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as temp_audio:
                # Convert to float32 and normalize
                audio_data = audio_data.astype(np.float32)
                audio_data = audio_data / np.max(np.abs(audio_data))
                
                # Whisper expects mono audio at 16kHz
                import scipy.io.wavfile as wavfile
                wavfile.write(temp_audio.name, 16000, audio_data)
                audio_path = temp_audio.name
        else:
            # File upload
            audio_path = audio
        
        # Transcribe
        print(f"Transcribing: {audio_path}")
        result = model.transcribe(audio_path, fp16=False)  # fp16=False for CPU
        
        # Clean up temp file if created
        if isinstance(audio, tuple):
            os.unlink(audio_path)
        
        transcription = result["text"]
        language = result["language"]
        
        print(f"Transcription complete. Language: {language}")
        
        return transcription, language
    
    except Exception as e:
        error_msg = f"Error during transcription: {str(e)}"
        print(error_msg)
        return error_msg, "Error"

# Create Gradio Interface
with gr.Blocks(title="Whisper ASR API") as demo:
    gr.Markdown(
        """
        # 🎙️ Whisper ASR - Speech to Text
        Upload an audio file or record your voice to get transcription.
        Supports multiple languages automatically!
        """
    )
    
    with gr.Tab("Upload Audio"):
        with gr.Row():
            with gr.Column():
                audio_input = gr.Audio(
                    sources=["upload"],
                    type="filepath",
                    label="Upload Audio File"
                )
                upload_button = gr.Button("Transcribe", variant="primary")
            
            with gr.Column():
                transcription_output = gr.Textbox(
                    label="Transcription",
                    lines=5,
                    placeholder="Transcription will appear here..."
                )
                language_output = gr.Textbox(
                    label="Detected Language",
                    lines=1
                )
        
        upload_button.click(
            fn=transcribe_audio,
            inputs=audio_input,
            outputs=[transcription_output, language_output]
        )
    
    with gr.Tab("Record Audio"):
        with gr.Row():
            with gr.Column():
                mic_input = gr.Audio(
                    sources=["microphone"],
                    type="numpy",
                    label="Record Audio"
                )
                record_button = gr.Button("Transcribe Recording", variant="primary")
            
            with gr.Column():
                mic_transcription_output = gr.Textbox(
                    label="Transcription",
                    lines=5,
                    placeholder="Transcription will appear here..."
                )
                mic_language_output = gr.Textbox(
                    label="Detected Language",
                    lines=1
                )
        
        record_button.click(
            fn=transcribe_audio,
            inputs=mic_input,
            outputs=[mic_transcription_output, mic_language_output]
        )
    
    gr.Markdown(
        """
        ### 📝 Notes:
        - Supports 99+ languages
        - Model: Whisper Base (fastest, good accuracy)
        - Processing time: ~5-15 seconds depending on audio length
        - Max audio length: 30 seconds recommended for best performance
        """
    )

# Launch the app
if __name__ == "__main__":
    demo.launch(
        server_name="0.0.0.0",
        server_port=7860,
        share=False
    )